Fig. 3

Performance characteristics of the “Protonomer” module for virus detection. RNA-Seq data from 24 samples known to harbor respiratory viruses (Additional file 1: Figure S9 and Table S11) were binned and the “viral” and “unclassified” bins were taxonomically classified by Protonomer, RAPSearch2 [36] (default and fast settings), and DIAMOND [37] (default and sensitive settings). Mean pairwise, genome-level sequence identities of the 24 respiratory viruses to reference sequences in the NCBI nt database were 93.7 % (range, 75.9–99.8 %). a Sensitivity. Protonomer (94.6 ± 2.7 %) and RAPSearch2 (default, 95.0 ± 2.2 %; fast, 94.8 ± 2.2 %) were more sensitive than DIAMOND (default, 90.5 ± 2.7 %; sensitive, 90.5 ± 2.7 %). b Specificity. Conversely, Protonomer (90.7 ± 17.1 %) and DIAMOND (default: 92.0 ± 17.1 %, sensitive: 91.9 ± 14.9 %) provided higher specificity than RAPSearch2 in default mode (88.0 ± 20.0 %). c Analysis times. Protonomer classifies reads faster than RAPSearch2 (24-fold compared to default mode, 11-fold compared to fast mode) and DIAMOND (2.6-fold compared to default mode, 3.3-fold compared to sensitive mode). All tools were run on 16 CPUs